ACPIPNP and too large IO resources

ACPIPNP and too large IO resources

[Pierre Ossman]

Hi there!

Commit 1acfb7f2b0d460ee86bdb25ad0679070ec8a5f0d by Bjorn is causing me
some grief. Although the patch seems correct, it is triggering another
misfeature of the system and I am hoping you have a solution.

Before your patch, the PCI bridge didn't allocate many io ports as they
were mislabeled as iomem. But now it puts its dirty paws all over the
entire ISA io port address space, effectively disabling PNP.

On my machine it steals the ranges 0x0-0xcf7, 0xcf8-0xcff and
0xd00-0xffff. IOW, the entire range of 0x0-0xffff gets blocked and none
of the ISA PNP devices can use ports outside this range.

We can see the same effect in the example given in your commit where
only the range 0x3b0-0x3df is left open.

I don't know enough about PNP to determine the problem, but I guess it's
the section that checks overlaps with other PNP devices that is somehow
wrong. It could also be that everyone keeps coding their DSDTs wrong,
but if that's the case then I see little other choice than to be bug
compatible.

Rgds
Pierre

Re: ACPIPNP and too large IO resources

[Bjorn Helgaas]

On Wednesday 05 July 2006 00:47, Pierre Ossman wrote:
> Commit 1acfb7f2b0d460ee86bdb25ad0679070ec8a5f0d by Bjorn is causing me
> some grief. Although the patch seems correct, it is triggering another
> misfeature of the system and I am hoping you have a solution.
> 
> Before your patch, the PCI bridge didn't allocate many io ports as they
> were mislabeled as iomem. But now it puts its dirty paws all over the
> entire ISA io port address space, effectively disabling PNP.
> 
> On my machine it steals the ranges 0x0-0xcf7, 0xcf8-0xcff and
> 0xd00-0xffff. IOW, the entire range of 0x0-0xffff gets blocked and none
> of the ISA PNP devices can use ports outside this range.

Thanks for the report!

It sounds like this might be the same problem as
    http://bugzilla.kernel.org/show_bug.cgi?id=6292

In short, you probably have a bridge device that consumes the
entire 0x0-0xffff I/O port range and produces some or all of that
range for downstream PNP devices.  PNP doesn't know what to do
with these windows that are both consumed by the bridge and made
available to downstream devices, so it just marks them as being
already reserved.

Matthieu Castet wrote a nice patch (attached) that makes PNP just
ignore those windows.  Can you try it and see whether it fixes
the problem you're seeing?  This patch is already in -mm, but not
yet in mainline.  We might need to consider this patch as
2.6.18 material if it resolves your problem.  I suspect many
people will see the same problem.

If it doesn't fix the problem, could I trouble you to open a
report at http://bugzilla.kernel.org and assign it to me?  If
you could attach the dmesg log and the output of "for i in
/sys/bus/pnp/devices/*; do echo $i; cat $i/id; cat $i/resources;
cat $i/options; done", that would be useful.

Thanks,
  Bjorn


Index: linux-2.6.16/drivers/pnp/pnpacpi/rsparser.c
===================================================================
--- linux-2.6.16.orig/drivers/pnp/pnpacpi/rsparser.c	2006-06-10 23:00:34.000000000 +0200
+++ linux-2.6.16/drivers/pnp/pnpacpi/rsparser.c	2006-06-10 23:08:52.114104816 +0200
@@ -170,6 +170,9 @@
 		return;
 	}
 
+    if (p->producer_consumer == ACPI_PRODUCER)
+        return;
+
 	if (p->resource_type == ACPI_MEMORY_RANGE)
 		pnpacpi_parse_allocated_memresource(res_table,
 				p->minimum, p->address_length);
@@ -248,9 +251,14 @@
 		break;
 
 	case ACPI_RESOURCE_TYPE_EXTENDED_ADDRESS64:
+        if (res->data.ext_address64.producer_consumer == ACPI_PRODUCER)
+            return AE_OK;
 		break;
 
 	case ACPI_RESOURCE_TYPE_EXTENDED_IRQ:
+        if (res->data.extended_irq.producer_consumer == ACPI_PRODUCER)
+            return AE_OK;
+
 		for (i = 0; i < res->data.extended_irq.interrupt_count; i++) {
 			pnpacpi_parse_allocated_irqresource(res_table,
 				res->data.extended_irq.interrupts[i],

Re: ACPIPNP and too large IO resources

[Pierre Ossman]

Bjorn Helgaas wrote:
> Thanks for the report!
>   

Np. Gave me an excuse to try out git bisect. ;)

> It sounds like this might be the same problem as
>     http://bugzilla.kernel.org/show_bug.cgi?id=6292
>
> In short, you probably have a bridge device that consumes the
> entire 0x0-0xffff I/O port range and produces some or all of that
> range for downstream PNP devices.  PNP doesn't know what to do
> with these windows that are both consumed by the bridge and made
> available to downstream devices, so it just marks them as being
> already reserved.
>   

Ah, that explains things.

> Matthieu Castet wrote a nice patch (attached) that makes PNP just
> ignore those windows.  Can you try it and see whether it fixes
> the problem you're seeing?  This patch is already in -mm, but not
> yet in mainline.  We might need to consider this patch as
> 2.6.18 material if it resolves your problem.  I suspect many
> people will see the same problem.
>   

The patch works nicely and removes all memory and io regions for the PCI
bridge but for the range 0xcf8-0xcff.

Thanks
Pierre

Re: ACPIPNP and too large IO resources

[Bjorn Helgaas]

On Wednesday 05 July 2006 14:53, Pierre Ossman wrote:
> Bjorn Helgaas wrote:
> > It sounds like this might be the same problem as
> >     http://bugzilla.kernel.org/show_bug.cgi?id=6292
> >
> > In short, you probably have a bridge device that consumes the
> > entire 0x0-0xffff I/O port range and produces some or all of that
> > range for downstream PNP devices.  PNP doesn't know what to do
> > with these windows that are both consumed by the bridge and made
> > available to downstream devices, so it just marks them as being
> > already reserved.
> 
> Ah, that explains things.
> 
> > Matthieu Castet wrote a nice patch (attached) that makes PNP just
> > ignore those windows.  Can you try it and see whether it fixes
> > the problem you're seeing?  This patch is already in -mm, but not
> > yet in mainline.  We might need to consider this patch as
> > 2.6.18 material if it resolves your problem.  I suspect many
> > people will see the same problem.
> 
> The patch works nicely and removes all memory and io regions for the PCI
> bridge but for the range 0xcf8-0xcff.

Andrew, I think we should try again to push
pnpacpi-reject-acpi_producer-resources.patch to the mainline.

Pierre's report (starts here: http://lkml.org/lkml/2006/7/5/20)
is another instance of http://bugzilla.kernel.org/show_bug.cgi?id=6292.

I suspect that many PNP devices are broken in 2.6.17 because of
this problem.  Probably the only reason we haven't seen more
reports is that PNPACPI isn't turned on by default.  (Maybe
we should do that in -mm?)

You recently proposed pushing it:
    http://marc.theaimsgroup.com/?l=linux-acpi&m=115119275408021&w=2
Len initially nacked it, but I think the outcome of the discussion
is that Shaohua doesn't object to this patch.  He probably would
still like to blacklist PNP0A03, but that's an additional step we
don't have to take at the same time.

Bjorn

Re: ACPIPNP and too large IO resources

[Andrew Morton]

Bjorn Helgaas <bjorn.helgaas@hp.com> wrote:
>
> On Wednesday 05 July 2006 14:53, Pierre Ossman wrote:
> > Bjorn Helgaas wrote:
> > > It sounds like this might be the same problem as
> > >     http://bugzilla.kernel.org/show_bug.cgi?id=6292
> > >
> > > In short, you probably have a bridge device that consumes the
> > > entire 0x0-0xffff I/O port range and produces some or all of that
> > > range for downstream PNP devices.  PNP doesn't know what to do
> > > with these windows that are both consumed by the bridge and made
> > > available to downstream devices, so it just marks them as being
> > > already reserved.
> > 
> > Ah, that explains things.
> > 
> > > Matthieu Castet wrote a nice patch (attached) that makes PNP just
> > > ignore those windows.  Can you try it and see whether it fixes
> > > the problem you're seeing?  This patch is already in -mm, but not
> > > yet in mainline.  We might need to consider this patch as
> > > 2.6.18 material if it resolves your problem.  I suspect many
> > > people will see the same problem.
> > 
> > The patch works nicely and removes all memory and io regions for the PCI
> > bridge but for the range 0xcf8-0xcff.
> 
> Andrew, I think we should try again to push
> pnpacpi-reject-acpi_producer-resources.patch to the mainline.
> 
> Pierre's report (starts here: http://lkml.org/lkml/2006/7/5/20)
> is another instance of http://bugzilla.kernel.org/show_bug.cgi?id=6292.
> 
> I suspect that many PNP devices are broken in 2.6.17 because of
> this problem.  Probably the only reason we haven't seen more
> reports is that PNPACPI isn't turned on by default.  (Maybe
> we should do that in -mm?)
> 
> You recently proposed pushing it:
>     http://marc.theaimsgroup.com/?l=linux-acpi&m=115119275408021&w=2
> Len initially nacked it, but I think the outcome of the discussion
> is that Shaohua doesn't object to this patch.  He probably would
> still like to blacklist PNP0A03, but that's an additional step we
> don't have to take at the same time.
> 

OK, well let's please push this up the priority list and work out what want
to do.  If Len's now OK with merging it then I think all lights are green?

pnpacpi-reject-acpi_producer-resources.patch:

From: matthieu castet <castet.matthieu@free.fr>

A patch in -mm kernel correct the parsing of "address resources" of pnpacpi. 
Before we assumed it was memory only, but it could be also IO.

But this change show an hidden bug : some resources could be producer type
that are not handled by pnp layer.  So we should ignore the producer
resources.

This patch fixes bug 6292 (http://bugzilla.kernel.org/show_bug.cgi?id=6292).
Some devices like PNP0A03 have 0xd00-0xffff and 0x0-0xcf7 as IO producer 
resources.

Before correcting "address resources" parsing, it was seen as memory and was
harmless, because nobody tried to reserve this memory range as it should be
IO.

With the correction it become IO resources, and make failed all others device
that want to register IO in this range and use pnp layer (like a ISA sound
card).

The solution is to ignore producer resources

Signed-off-by: Matthieu CASTET <castet.matthieu@free.fr>
Signed-off-by: Uwe Bugla <uwe.bugla@gmx.de>
Cc: Bjorn Helgaas <bjorn.helgaas@hp.com>
Cc: Adam Belay <ambx1@neo.rr.com>
Cc: "Brown, Len" <len.brown@intel.com>

akpm: previously nacked, as per comment #26.  But am hanging onto it until the
thing gets fixed for real.

Signed-off-by: Andrew Morton <akpm@osdl.org>
---

 drivers/pnp/pnpacpi/rsparser.c |    8 ++++++++
 1 file changed, 8 insertions(+)

diff -puN drivers/pnp/pnpacpi/rsparser.c~pnpacpi-reject-acpi_producer-resources drivers/pnp/pnpacpi/rsparser.c
--- a/drivers/pnp/pnpacpi/rsparser.c~pnpacpi-reject-acpi_producer-resources
+++ a/drivers/pnp/pnpacpi/rsparser.c
@@ -173,6 +173,9 @@ pnpacpi_parse_allocated_address_space(st
 		return;
 	}
 
+	if (p->producer_consumer == ACPI_PRODUCER)
+		return;
+
 	if (p->resource_type == ACPI_MEMORY_RANGE)
 		pnpacpi_parse_allocated_memresource(res_table,
 				p->minimum, p->address_length);
@@ -252,9 +255,14 @@ static acpi_status pnpacpi_allocated_res
 		break;
 
 	case ACPI_RESOURCE_TYPE_EXTENDED_ADDRESS64:
+		if (res->data.ext_address64.producer_consumer == ACPI_PRODUCER)
+			return AE_OK;
 		break;
 
 	case ACPI_RESOURCE_TYPE_EXTENDED_IRQ:
+		if (res->data.extended_irq.producer_consumer == ACPI_PRODUCER)
+			return AE_OK;
+
 		for (i = 0; i < res->data.extended_irq.interrupt_count; i++) {
 			pnpacpi_parse_allocated_irqresource(res_table,
 				res->data.extended_irq.interrupts[i],
_

Re: ACPIPNP and too large IO resources

[Bjorn Helgaas]

On Wednesday 05 July 2006 16:18, Andrew Morton wrote:
> Bjorn Helgaas <bjorn.helgaas@hp.com> wrote:
> > You recently proposed pushing it:
> >     http://marc.theaimsgroup.com/?l=linux-acpi&m=115119275408021&w=2
> > Len initially nacked it, but I think the outcome of the discussion
> > is that Shaohua doesn't object to this patch.  He probably would
> > still like to blacklist PNP0A03, but that's an additional step we
> > don't have to take at the same time.
> 
> OK, well let's please push this up the priority list and work out what want
> to do.  If Len's now OK with merging it then I think all lights are green?

My opinion is that we don't need to blacklist PNP0A03 unless
we discover a problem.  

Len, Shaohua, can you ack this?

> pnpacpi-reject-acpi_producer-resources.patch:
> 
> From: matthieu castet <castet.matthieu@free.fr>
> 
> A patch in -mm kernel correct the parsing of "address resources" of pnpacpi. 
> Before we assumed it was memory only, but it could be also IO.
> 
> But this change show an hidden bug : some resources could be producer type
> that are not handled by pnp layer.  So we should ignore the producer
> resources.
> 
> This patch fixes bug 6292 (http://bugzilla.kernel.org/show_bug.cgi?id=6292).
> Some devices like PNP0A03 have 0xd00-0xffff and 0x0-0xcf7 as IO producer 
> resources.
> 
> Before correcting "address resources" parsing, it was seen as memory and was
> harmless, because nobody tried to reserve this memory range as it should be
> IO.
> 
> With the correction it become IO resources, and make failed all others device
> that want to register IO in this range and use pnp layer (like a ISA sound
> card).
> 
> The solution is to ignore producer resources
> 
> Signed-off-by: Matthieu CASTET <castet.matthieu@free.fr>
> Signed-off-by: Uwe Bugla <uwe.bugla@gmx.de>
> Cc: Bjorn Helgaas <bjorn.helgaas@hp.com>
> Cc: Adam Belay <ambx1@neo.rr.com>
> Cc: "Brown, Len" <len.brown@intel.com>
> 
> akpm: previously nacked, as per comment #26.  But am hanging onto it until the
> thing gets fixed for real.
> 
> Signed-off-by: Andrew Morton <akpm@osdl.org>
> ---
> 
>  drivers/pnp/pnpacpi/rsparser.c |    8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff -puN drivers/pnp/pnpacpi/rsparser.c~pnpacpi-reject-acpi_producer-resources drivers/pnp/pnpacpi/rsparser.c
> --- a/drivers/pnp/pnpacpi/rsparser.c~pnpacpi-reject-acpi_producer-resources
> +++ a/drivers/pnp/pnpacpi/rsparser.c
> @@ -173,6 +173,9 @@ pnpacpi_parse_allocated_address_space(st
>  		return;
>  	}
>  
> +	if (p->producer_consumer == ACPI_PRODUCER)
> +		return;
> +
>  	if (p->resource_type == ACPI_MEMORY_RANGE)
>  		pnpacpi_parse_allocated_memresource(res_table,
>  				p->minimum, p->address_length);
> @@ -252,9 +255,14 @@ static acpi_status pnpacpi_allocated_res
>  		break;
>  
>  	case ACPI_RESOURCE_TYPE_EXTENDED_ADDRESS64:
> +		if (res->data.ext_address64.producer_consumer == ACPI_PRODUCER)
> +			return AE_OK;
>  		break;
>  
>  	case ACPI_RESOURCE_TYPE_EXTENDED_IRQ:
> +		if (res->data.extended_irq.producer_consumer == ACPI_PRODUCER)
> +			return AE_OK;
> +
>  		for (i = 0; i < res->data.extended_irq.interrupt_count; i++) {
>  			pnpacpi_parse_allocated_irqresource(res_table,
>  				res->data.extended_irq.interrupts[i],
> _
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

Re: ACPIPNP and too large IO resources

[Shaohua Li]

On Thu, 2006-07-06 at 23:29 +0800, Bjorn Helgaas wrote:
> On Wednesday 05 July 2006 16:18, Andrew Morton wrote: 
> > Bjorn Helgaas <bjorn.helgaas@hp.com> wrote: 
> > > You recently proposed pushing it: 
> > >
> http://marc.theaimsgroup.com/?l=linux-acpi&m=115119275408021&w=2 
> > > Len initially nacked it, but I think the outcome of the
> discussion 
> > > is that Shaohua doesn't object to this patch.  He probably would 
> > > still like to blacklist PNP0A03, but that's an additional step we 
> > > don't have to take at the same time. 
> >  
> > OK, well let's please push this up the priority list and work out
> what want 
> > to do.  If Len's now OK with merging it then I think all lights are
> green?
> 
> My opinion is that we don't need to blacklist PNP0A03 unless 
> we discover a problem.  
> 
> Len, Shaohua, can you ack this?
Sure. Please merge the patch Bjorn pointed out.

Thanks,
Shaohua

> 
> > pnpacpi-reject-acpi_producer-resources.patch: 
> >  
> > From: matthieu castet <castet.matthieu@free.fr> 
> >  
> > A patch in -mm kernel correct the parsing of "address resources" of
> pnpacpi.  
> > Before we assumed it was memory only, but it could be also IO. 
> >  
> > But this change show an hidden bug : some resources could be
> producer type 
> > that are not handled by pnp layer.  So we should ignore the
> producer 
> > resources. 
> >  
> > This patch fixes bug 6292
> (http://bugzilla.kernel.org/show_bug.cgi?id=6292). 
> > Some devices like PNP0A03 have 0xd00-0xffff and 0x0-0xcf7 as IO
> producer  
> > resources. 
> >  
> > Before correcting "address resources" parsing, it was seen as memory
> and was 
> > harmless, because nobody tried to reserve this memory range as it
> should be 
> > IO. 
> >  
> > With the correction it become IO resources, and make failed all
> others device 
> > that want to register IO in this range and use pnp layer (like a ISA
> sound 
> > card). 
> >  
> > The solution is to ignore producer resources 
> >  
> > Signed-off-by: Matthieu CASTET <castet.matthieu@free.fr> 
> > Signed-off-by: Uwe Bugla <uwe.bugla@gmx.de> 
> > Cc: Bjorn Helgaas <bjorn.helgaas@hp.com> 
> > Cc: Adam Belay <ambx1@neo.rr.com> 
> > Cc: "Brown, Len" <len.brown@intel.com> 
> >  
> > akpm: previously nacked, as per comment #26.  But am hanging onto it
> until the 
> > thing gets fixed for real. 
> >  
> > Signed-off-by: Andrew Morton <akpm@osdl.org> 
> > --- 
> >  
> >  drivers/pnp/pnpacpi/rsparser.c |    8 ++++++++ 
> >  1 file changed, 8 insertions(+) 
> >  
> > diff -puN
> drivers/pnp/pnpacpi/rsparser.c~pnpacpi-reject-acpi_producer-resources
> drivers/pnp/pnpacpi/rsparser.c 
> > ---
> a/drivers/pnp/pnpacpi/rsparser.c~pnpacpi-reject-acpi_producer-resources 
> > +++ a/drivers/pnp/pnpacpi/rsparser.c 
> > @@ -173,6 +173,9 @@ pnpacpi_parse_allocated_address_space(st 
> >               return; 
> >       } 
> >   
> > +     if (p->producer_consumer == ACPI_PRODUCER) 
> > +             return; 
> > + 
> >       if (p->resource_type == ACPI_MEMORY_RANGE) 
> >               pnpacpi_parse_allocated_memresource(res_table, 
> >                               p->minimum, p->address_length); 
> > @@ -252,9 +255,14 @@ static acpi_status pnpacpi_allocated_res 
> >               break; 
> >   
> >       case ACPI_RESOURCE_TYPE_EXTENDED_ADDRESS64: 
> > +             if (res->data.ext_address64.producer_consumer ==
> ACPI_PRODUCER) 
> > +                     return AE_OK; 
> >               break; 
> >   
> >       case ACPI_RESOURCE_TYPE_EXTENDED_IRQ: 
> > +             if (res->data.extended_irq.producer_consumer ==
> ACPI_PRODUCER) 
> > +                     return AE_OK; 
> > + 
> >               for (i = 0; i <
> res->data.extended_irq.interrupt_count; i++) { 
> >
> pnpacpi_parse_allocated_irqresource(res_table, 
> >                               res->data.extended_irq.interrupts[i], 
> > _ 
> >  
> > - 
> > To unsubscribe from this list: send the line "unsubscribe
> linux-acpi" in 
> > the body of a message to majordomo@vger.kernel.org 
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html 
> > 
> 

Re: ACPIPNP and too large IO resources

[Pierre Ossman]

Shaohua Li wrote:
> Sure. Please merge the patch Bjorn pointed out.
>   

So? What's the status of this? Still not in Linus' tree.

Rgds
Pierre

Re: ACPIPNP and too large IO resources

[Andrew Morton]

On Sat, 05 Aug 2006 02:19:46 +0200
Pierre Ossman <drzeus-list@drzeus.cx> wrote:

> Shaohua Li wrote:
> > Sure. Please merge the patch Bjorn pointed out.
> >   
> 
> So? What's the status of this? Still not in Linus' tree.

We have an ack from Shaohua.  I'll merge it.